{"id":"https://openalex.org/W7203701481","doi":"https://doi.org/10.48550/arxiv.2608.16333","title":"Step-Level On-Policy Distillation: Interpolating Between On-Policy Distillation and Supervised Fine-Tuning","display_name":"Step-Level On-Policy Distillation: Interpolating Between On-Policy Distillation and Supervised Fine-Tuning","publication_year":2026,"publication_date":"2026-08-17","ids":{"openalex":"https://openalex.org/W7203701481","doi":"https://doi.org/10.48550/arxiv.2608.16333"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2608.16333","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.16333","pdf_url":"https://arxiv.org/pdf/2608.16333","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2608.16333","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5122070121","display_name":"Changhui Sun","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Sun, Changhui","raw_affiliation_strings":["State Key Laboratory for Novel Software Technology , Nanjing University , China","XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"State Key Laboratory for Novel Software Technology , Nanjing University , China","institution_ids":["https://openalex.org/I881766915"]},{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122930430","display_name":"Lanbo Liu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Lanbo","raw_affiliation_strings":["XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5144396554","display_name":"Hang Lei","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Lei, Hang","raw_affiliation_strings":["XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5144416352","display_name":"Tong Ling","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210165038","display_name":"University of Chinese Academy of Sciences","ror":"https://ror.org/05qbk4x57","country_code":"CN","type":"education","lineage":["https://openalex.org/I19820366","https://openalex.org/I4210165038"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Ling, Tong","raw_affiliation_strings":["University of Chinese Academy of Sciences , Beijing , China","XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Chinese Academy of Sciences , Beijing , China","institution_ids":["https://openalex.org/I4210165038"]},{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5066972151","display_name":"Jiahang Xie","orcid":"https://orcid.org/0000-0003-0860-4733","observed_orcids":["https://orcid.org/0000-0003-0860-4733"]},"institutions":[{"id":"https://openalex.org/I24943067","display_name":"Fudan University","ror":"https://ror.org/013q1eq08","country_code":"CN","type":"education","lineage":["https://openalex.org/I24943067"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xie, Jiahang","raw_affiliation_strings":["School of Data Science , Fudan University , Shanghai , China","XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Data Science , Fudan University , Shanghai , China","institution_ids":["https://openalex.org/I24943067"]},{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5146594088","display_name":"Zhiyong Zheng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zheng, Zhiyong","raw_affiliation_strings":["XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5147391932","display_name":"Yujia Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Yujia","raw_affiliation_strings":["XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5147443536","display_name":"Hao Liu","orcid":"https://orcid.org/0000-0002-1718-1448","observed_orcids":["https://orcid.org/0000-0002-1718-1448"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Hao","raw_affiliation_strings":["XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122948652","display_name":"Feng Xiao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiao, Feng","raw_affiliation_strings":["XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5122327742","display_name":"Lu Liu","orcid":"https://orcid.org/0000-0002-9376-1487","observed_orcids":["https://orcid.org/0000-0002-9376-1487","https://orcid.org/0000-0002-1613-4287","https://orcid.org/0000-0003-4658-6361","https://orcid.org/0000-0002-6942-1095"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Lu","raw_affiliation_strings":["XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5147405079","display_name":"Yanlong Du","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Du, Yanlong","raw_affiliation_strings":["XingYun Lab , HUJING Digital Media & Entertainment Group"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"XingYun Lab , HUJING Digital Media & Entertainment Group","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5033620369","display_name":"Zifeng Cheng","orcid":"https://orcid.org/0000-0002-8486-2614","observed_orcids":["https://orcid.org/0000-0002-8486-2614"]},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Cheng, Zifeng","raw_affiliation_strings":["State Key Laboratory for Novel Software Technology , Nanjing University , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"State Key Laboratory for Novel Software Technology , Nanjing University , China","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5036663072","display_name":"Ziwei Jiang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jiang, Ziwei","raw_affiliation_strings":["State Key Laboratory for Novel Software Technology , Nanjing University , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"State Key Laboratory for Novel Software Technology , Nanjing University , China","institution_ids":["https://openalex.org/I881766915"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5140767239","display_name":"Qing Gu","orcid":"https://orcid.org/0000-0002-1112-790X","observed_orcids":["https://orcid.org/0000-0002-1112-790X","https://orcid.org/0009-0007-8493-7464"]},"institutions":[{"id":"https://openalex.org/I881766915","display_name":"Nanjing University","ror":"https://ror.org/01rxvg760","country_code":"CN","type":"education","lineage":["https://openalex.org/I881766915"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Gu, Qing","raw_affiliation_strings":["State Key Laboratory for Novel Software Technology , Nanjing University , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"State Key Laboratory for Novel Software Technology , Nanjing University , China","institution_ids":["https://openalex.org/I881766915"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":3,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.975600004196167,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.975600004196167,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.005200000014156103,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10906","display_name":"AI-based Problem Solving and Planning","score":0.003599999938160181,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/on-policy-distillation","display_name":"on-policy distillation","score":0.9909999966621399},{"id":"https://openalex.org/keywords/supervised-fine-tuning","display_name":"supervised fine-tuning","score":0.9800000190734863},{"id":"https://openalex.org/keywords/knowledge-distillation","display_name":"knowledge distillation","score":0.7599999904632568},{"id":"https://openalex.org/keywords/teacher-student-alignment","display_name":"teacher-student alignment","score":0.703000009059906},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.5490000247955322},{"id":"https://openalex.org/keywords/reasoning-tasks","display_name":"reasoning tasks","score":0.4779999852180481},{"id":"https://openalex.org/keywords/alfworld","display_name":"ALFWorld","score":0.7760000228881836}],"concepts":[{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.6600000262260437},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.6514999866485596},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5273000001907349},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.5246000289916992},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.44999998807907104},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.4230000078678131},{"id":"https://openalex.org/C18762648","wikidata":"https://www.wikidata.org/wiki/Q42213","display_name":"Work (physics)","level":2,"score":0.4212000072002411},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.3937999904155731},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3517000079154968},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.30489999055862427},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.30390000343322754},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.2653000056743622},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.2587999999523163},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.25450000166893005},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.2542000114917755}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2608.16333","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.16333","pdf_url":"https://arxiv.org/pdf/2608.16333","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2608.16333","doi":"https://doi.org/10.48550/arxiv.2608.16333","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.16333","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2608.16333","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.16333","pdf_url":"https://arxiv.org/pdf/2608.16333","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[{"score":0.5,"display_name":"Quality education","id":"https://metadata.un.org/sdg/4"}],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7203701481.pdf","grobid_xml":"https://content.openalex.org/works/W7203701481.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7204908525","https://openalex.org/W7162992044","https://openalex.org/W7163208236","https://openalex.org/W7222327894","https://openalex.org/W7167615039","https://openalex.org/W7163595920","https://openalex.org/W7214777665","https://openalex.org/W7141388625","https://openalex.org/W7214953522","https://openalex.org/W7223059966"],"abstract_inverted_index":{"On-policy":[0],"distillation":[1,28,179],"(OPD)":[2],"aligns":[3],"a":[4,8,50,172],"student":[5,45,116],"model":[6],"with":[7,29,75,106,123,127],"teacher's":[9],"logit":[10],"distribution":[11],"on":[12,115,153,178],"student-generated":[13,87],"trajectories.":[14,88],"This":[15],"approach":[16],"has":[17],"achieved":[18],"strong":[19],"empirical":[20],"gains":[21],"and":[22,47,52,118,141,149],"can":[23,37],"often":[24],"surpass":[25],"conventional":[26,147],"off-policy":[27],"substantially":[30,145],"less":[31],"data.":[32],"However,":[33],"standard":[34],"token-level":[35,136],"OPD":[36,80],"provide":[38,82],"only":[39],"fragmented":[40,135],"corrections":[41,132],"along":[42],"an":[43],"erroneous":[44],"trajectory":[46],"cannot":[48],"unfold":[49],"complete":[51,86],"correct":[53],"repair":[54],"path.":[55],"Motivated":[56],"by":[57,161],"this":[58,169],"limitation,":[59],"we":[60],"propose":[61],"\\emph{Step-Level":[62],"On-Policy":[63],"Distillation}":[64],"(SOPD),":[65],"which":[66],"combines":[67],"the":[68,76,108,157],"long-horizon":[69],"correction":[70],"of":[71,79,95],"supervised":[72],"fine-tuning":[73],"(SFT)":[74],"on-policy":[77],"advantage":[78],"to":[81,100],"step-level":[83],"supervision":[84],"over":[85,164],"We":[89,167],"show":[90],"that,":[91],"at":[92],"different":[93],"limits":[94],"step":[96],"length,":[97],"SOPD":[98,112,129,144,155],"reduces":[99],"SFT":[101,148],"or":[102],"approximates":[103],"OPD.":[104,150,166],"Compared":[105],"SFT,":[107],"teacher":[109],"responses":[110],"in":[111],"are":[113],"conditioned":[114],"trajectories":[117],"therefore":[119],"align":[120],"more":[121],"closely":[122],"student-visited":[124],"states;":[125],"compared":[126],"OPD,":[128],"provides":[130],"longer-horizon":[131],"rather":[133],"than":[134],"guidance.":[137],"Across":[138],"both":[139],"reasoning":[140],"agent":[142],"tasks,":[143],"outperforms":[146],"For":[151],"example,":[152],"ALFWorld,":[154],"improves":[156],"average":[158],"success":[159],"rate":[160],"13.4":[162],"points":[163],"Vanilla":[165],"hope":[168],"work":[170],"offers":[171],"new":[173],"perspective":[174],"for":[175],"future":[176],"research":[177],"methods.":[180]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-08-19T00:00:00"}